feat: add researched fixed companion builds #1

• 
  @anthodev opened this merge request · Sep 26, 2026

  Closes #1.

  What

  • Canonical curated builds for all 22 non-Cross companions in new static data src/data/companionBuilds.ts: ordered Arts/Skills, bilingual role/rationale, character-specific sourced weapon/armor/availability advice, and source links with DE/XCX/both edition badges.
  • Strict companion TP ceiling: at most 2 TP-consuming Arts per companion loadout, including substitutes; Cross keeps its own 3-Art budget (TP_ART_BUDGET) unchanged.
  • Deterministic progression substitutes: unavailable canonical abilities fall back to explicit, ordered personal-rank substitutes — never dynamic strategy re-ranking. Canonical Arts reclaim their slot as they unlock (regression-tested, e.g. Phog Side Slash at rank 5).
  • Fixed kits invariant to Cross class, weapons, teammates, team variant and combat mode; Skell mode no longer substitutes ground Skills (planCompanionForMode removed).
  • Equipped-only companion scoring: evaluateCompanionContext scores the equipped Arts/Skills; Skell aptitude counts equipped pilot Skills only — no credit for unlocked reserves.
  • BuildCard UI (EN/FR): RESEARCHED BUILD badge, curated role, build rationale, ordered Art/Skill substitute ladders, per-character equipment advice (generic strategy manufacturer/stat blocks suppressed for curated companions), sources with edition badges. Skell packages remain separate.
  • Research matrices (docs/research/companion-builds-{core,veterans,specialists,remaining}.md): per-Art personal rank, weapon, TP flag and source URL; documents guide errors rejected (weapon-incompatible lists, missing TP markers, invalid Art names), sparse-evidence editorial choices, and explicit disagreements (e.g. Iron Saber attribute).

  Evidence

  • 57/57 tests pass (fixedCompanionBuilds.test.ts: roster coverage, legality/weapon triple-check, TP ceiling, kit invariance across 8 Cross configs, determinism, rank clamps, progression boundaries; contextualEngine equipped-only tests).
  • typecheck, eslint app src, production build clean. Repo-wide npm run lint still reports pre-existing errors in .claude/helpers/*.cjs (out of scope).
  • 24-case Cross baseline (6 final classes × signature flag × legacy/contextual) byte-identical before/after.
  • Production UI smoke EN/FR: fixed kits across class/mode/variant changes, Phog rank 4→5 substitution, FR equipment advice, keyboard-operable source disclosures, no console errors, no horizontal overflow desktop/mobile.

  Rickub deployment caveat

  Rickub ignores environment: production: it does not enforce environment reviewers, protection rules or wait timers. The Bunny workflow still deploys automatically on pushes to develop after its checks pass when its deployment secrets are configured. Any production-environment approvals on the previous host are not carried over; their prior configuration was not verified.

  Deliberate choices

  • Irina equips 3 of 4 Skill slots: her remaining personal Skills depend on Sleep/Control the curated support bar never provides; an inert fourth Skill is worse than an empty slot.
  • Affinity missions teach companion signature Arts to Cross only; they are not an extra gate on the owner's personal-rank Art.

  🤖 Generated with omp

  edited

Sign in to comment.